PyTorchKR Papers

최신 논문

총 6,378편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2505.05229

Does CLIP perceive art the same way we do?

CLIP has emerged as a powerful multimodal model capable of connecting images and text through joint embeddings, but to what extent does it "see" the same way humans do - especially when interpreting artworks? In this paper, we investigate CLIP's ability to extract high-level semantic and stylistic information from paintings, including both human-created and AI-generated imagery. We evaluate its perception across multiple dimensions: content, scene understanding, artistic style, historical period, and the presence of visual deformations or artifacts. By designing targeted probing tasks and comparing CLIP's responses to human annotations and expert benchmarks, we explore its alignment with human perceptual and contextual understanding. Our findings reveal both strengths and limitations in CLIP's visual representations, particularly in relation to aesthetic cues and artistic intent. We further discuss the implications of these insights for using CLIP as a guidance mechanism during generative processes, such as style transfer or prompt-based image synthesis. Our work highlights the need for deeper interpretability in multimodal systems, especially when applied to creative domains where nuance and subjectivity play a central role.

arXiv:2505.05081

PIDiff: Image Customization for Personalized Identities with Diffusion Models

Text-to-image generation for personalized identities aims at incorporating the specific identity into images using a text prompt and an identity image. Based on the powerful generative capabilities of DDPMs, many previous works adopt additional prompts, such as text embeddings and CLIP image embeddings, to represent the identity information, while they fail to disentangle the identity information and background information. As a result, the generated images not only lose key identity characteristics but also suffer from significantly reduced diversity. To address this issue, previous works have combined the W+ space from StyleGAN with diffusion models, leveraging this space to provide a more accurate and comprehensive representation of identity features through multi-level feature extraction. However, the entanglement of identity and background information in in-the-wild images during training prevents accurate identity localization, resulting in severe semantic interference between identity and background. In this paper, we propose a novel fine-tuning-based diffusion model for personalized identities text-to-image generation, named PIDiff, which leverages the W+ space and an identity-tailored fine-tuning strategy to avoid semantic entanglement and achieves accurate feature extraction and localization. Style editing can also be achieved by PIDiff through preserving the characteristics of identity features in the W+ space, which vary from coarse to fine. Through the combination of the proposed cross-attention block and parameter optimization strategy, PIDiff preserves the identity information and maintains the generation capability for in-the-wild images of the pre-trained model during inference. Our experimental results validate the effectiveness of our method in this task.

arXiv:2505.04851

CRAFT: Cultural Russian-Oriented Dataset Adaptation for Focused Text-to-Image Generation

Despite the fact that popular text-to-image generation models cope well with international and general cultural queries, they have a significant knowledge gap regarding individual cultures. This is due to the content of existing large training datasets collected on the Internet, which are predominantly based on Western European or American popular culture. Meanwhile, the lack of cultural adaptation of the model can lead to incorrect results, a decrease in the generation quality, and the spread of stereotypes and offensive content. In an effort to address this issue, we examine the concept of cultural code and recognize the critical importance of its understanding by modern image generation models, an issue that has not been sufficiently addressed in the research community to date. We propose the methodology for collecting and processing the data necessary to form a dataset based on the cultural code, in particular the Russian one. We explore how the collected data affects the quality of generations in the national domain and analyze the effectiveness of our approach using the Kandinsky 3.1 text-to-image model. Human evaluation results demonstrate an increase in the level of awareness of Russian culture in the model.

arXiv:2505.04650공식 코드 · Papers with Code

Multimodal Benchmarking and Recommendation of Text-to-Image Generation Models

This work presents an open-source unified benchmarking and evaluation framework for text-to-image generation models, with a particular focus on the impact of metadata augmented prompts. Leveraging the DeepFashion-MultiModal dataset, we assess generated outputs through a comprehensive set of quantitative metrics, including Weighted Score, CLIP (Contrastive Language Image Pre-training)-based similarity, LPIPS (Learned Perceptual Image Patch Similarity), FID (Frechet Inception Distance), and retrieval-based measures, as well as qualitative analysis. Our results demonstrate that structured metadata enrichments greatly enhance visual realism, semantic fidelity, and model robustness across diverse text-to-image architectures. While not a traditional recommender system, our framework enables task-specific recommendations for model selection and prompt design based on evaluation metrics.

arXiv:2504.10146공식 코드 · Papers with Code

GeoUni: A Unified Model for Generating Geometry Diagrams, Problems and Problem Solutions

We propose GeoUni, the first unified geometry expert model capable of generating problem solutions and diagrams within a single framework in a way that enables the creation of unique and individualized geometry problems. Traditionally, solving geometry problems and generating diagrams have been treated as separate tasks in machine learning, with no models successfully integrating both to support problem creation. However, we believe that mastery in geometry requires frictionless integration of all of these skills, from solving problems to visualizing geometric relationships, and finally, crafting tailored problems. Our extensive experiments demonstrate that GeoUni, with only 1.5B parameters, achieves performance comparable to larger models such as DeepSeek-R1 with 671B parameters in geometric reasoning tasks. GeoUni also excels in generating precise geometric diagrams, surpassing both text-to-image models and unified models, including the GPT-4o image generation. Most importantly, GeoUni is the only model capable of successfully generating textual problems with matching diagrams based on specific knowledge points, thus offering a wider range of capabilities that extend beyond current models.

arXiv:2505.04347

CountDiffusion: Text-to-Image Synthesis with Training-Free Counting-Guidance Diffusion

Stable Diffusion has advanced text-to-image synthesis, but training models to generate images with accurate object quantity is still difficult due to the high computational cost and the challenge of teaching models the abstract concept of quantity. In this paper, we propose CountDiffusion, a training-free framework aiming at generating images with correct object quantity from textual descriptions. CountDiffusion consists of two stages. In the first stage, an intermediate denoising result is generated by the diffusion model to predict the final synthesized image with one-step denoising, and a counting model is used to count the number of objects in this image. In the second stage, a correction module is used to correct the object quantity by changing the attention map of the object with universal guidance. The proposed CountDiffusion can be plugged into any diffusion-based text-to-image (T2I) generation models without further training. Experiment results demonstrate the superiority of our proposed CountDiffusion, which improves the accurate object quantity generation ability of T2I models by a large margin.

arXiv:2504.21814

Why Compress What You Can Generate? When GPT-4o Generation Ushers in Image Compression Fields

The rapid development of AIGC foundation models has revolutionized the paradigm of image compression, which paves the way for the abandonment of most pixel-level transform and coding, compelling us to ask: why compress what you can generate if the AIGC foundation model is powerful enough to faithfully generate intricate structure and fine-grained details from nothing more than some compact descriptors, i.e., texts, or cues. Fortunately, recent GPT-4o image generation of OpenAI has achieved impressive cross-modality generation, editing, and design capabilities, which motivates us to answer the above question by exploring its potential in image compression fields. In this work, we investigate two typical compression paradigms: textual coding and multimodal coding (i.e., text + extremely low-resolution image), where all/most pixel-level information is generated instead of compressing via the advanced GPT-4o image generation function. The essential challenge lies in how to maintain semantic and structure consistency during the decoding process. To overcome this, we propose a structure raster-scan prompt engineering mechanism to transform the image into textual space, which is compressed as the condition of GPT-4o image generation. Extensive experiments have shown that the combination of our designed structural raster-scan prompts and GPT-4o's image generation function achieved the impressive performance compared with recent multimodal/generative image compression at ultra-low bitrate, further indicating the potential of AIGC generation in image compression fields.

arXiv:2504.17502

RefVNLI: Towards Scalable Evaluation of Subject-driven Text-to-image Generation

Subject-driven text-to-image (T2I) generation aims to produce images that align with a given textual description, while preserving the visual identity from a referenced subject image. Despite its broad downstream applicability -- ranging from enhanced personalization in image generation to consistent character representation in video rendering -- progress in this field is limited by the lack of reliable automatic evaluation. Existing methods either assess only one aspect of the task (i.e., textual alignment or subject preservation), misalign with human judgments, or rely on costly API-based evaluation. To address this, we introduce RefVNLI, a cost-effective metric that evaluates both textual alignment and subject preservation in a single prediction. Trained on a large-scale dataset derived from video-reasoning benchmarks and image perturbations, RefVNLI outperforms or matches existing baselines across multiple benchmarks and subject categories (e.g., \emph{Animal}, \emph{Object}), achieving up to 6.4-point gains in textual alignment and 8.5-point gains in subject consistency. It also excels with lesser-known concepts, aligning with human preferences at over 87\% accuracy.

arXiv:2504.00999공식 코드 · Papers with Code

MergeVQ: A Unified Framework for Visual Generation and Representation with Disentangled Token Merging and Quantization

Masked Image Modeling (MIM) with Vector Quantization (VQ) has achieved great success in both self-supervised pre-training and image generation. However, most existing methods struggle to address the trade-off in shared latent space for generation quality vs. representation learning and efficiency. To push the limits of this paradigm, we propose MergeVQ, which incorporates token merging techniques into VQ-based generative models to bridge the gap between image generation and visual representation learning in a unified architecture. During pre-training, MergeVQ decouples top-k semantics from latent space with the token merge module after self-attention blocks in the encoder for subsequent Look-up Free Quantization (LFQ) and global alignment and recovers their fine-grained details through cross-attention in the decoder for reconstruction. As for the second-stage generation, we introduce MergeAR, which performs KV Cache compression for efficient raster-order prediction. Extensive experiments on ImageNet verify that MergeVQ as an AR generative model achieves competitive performance in both visual representation learning and image generation tasks while maintaining favorable token efficiency and inference speed. The code and model will be available at https://apexgen-x.github.io/MergeVQ.

arXiv:2505.04105

MAISY: Motion-Aware Image SYnthesis for Medical Image Motion Correction

Patient motion during medical image acquisition causes blurring, ghosting, and distorts organs, which makes image interpretation challenging. Current state-of-the-art algorithms using Generative Adversarial Network (GAN)-based methods with their ability to learn the mappings between corrupted images and their ground truth via Structural Similarity Index Measure (SSIM) loss effectively generate motion-free images. However, we identified the following limitations: (i) they mainly focus on global structural characteristics and therefore overlook localized features that often carry critical pathological information, and (ii) the SSIM loss function struggles to handle images with varying pixel intensities, luminance factors, and variance. In this study, we propose Motion-Aware Image SYnthesis (MAISY) which initially characterize motion and then uses it for correction by: (a) leveraging the foundation model Segment Anything Model (SAM), to dynamically learn spatial patterns along anatomical boundaries where motion artifacts are most pronounced and, (b) introducing the Variance-Selective SSIM (VS-SSIM) loss which adaptively emphasizes spatial regions with high pixel variance to preserve essential anatomical details during artifact correction. Experiments on chest and head CT datasets demonstrate that our model outperformed the state-of-the-art counterparts, with Peak Signal-to-Noise Ratio (PSNR) increasing by 40%, SSIM by 10%, and Dice by 16%.

arXiv:2505.04497

Defining and Quantifying Creative Behavior in Popular Image Generators

Creativity of generative AI models has been a subject of scientific debate in the last years, without a conclusive answer. In this paper, we study creativity from a practical perspective and introduce quantitative measures that help the user to choose a suitable AI model for a given task. We evaluated our measures on a number of popular image-to-image generation models, and the results of this suggest that our measures conform to human intuition.

arXiv:2505.04486

Efficient Flow Matching using Latent Variables

Flow matching models have shown great potential in image generation tasks among probabilistic generative models. Building upon the ideas of continuous normalizing flows, flow matching models generalize the transport path of the diffusion models from a simple prior distribution to the data. Most flow matching models in the literature do not explicitly model the underlying structure/manifold in the target data when learning the flow from a simple source distribution like the standard Gaussian. This leads to inefficient learning, especially for many high-dimensional real-world datasets, which often reside in a low-dimensional manifold. Existing strategies of incorporating manifolds, including data with underlying multi-modal distribution, often require expensive training and hence frequently lead to suboptimal performance. To this end, we present \texttt{Latent-CFM}, which provides simplified training/inference strategies to incorporate multi-modal data structures using pretrained deep latent variable models. Through experiments on multi-modal synthetic data and widely used image benchmark datasets, we show that \texttt{Latent-CFM} exhibits improved generation quality with significantly less training ($\sim 50\%$ less in some cases) and computation than state-of-the-art flow matching models. Using a 2d Darcy flow dataset, we demonstrate that our approach generates more physically accurate samples than competitive approaches. In addition, through latent space analysis, we demonstrate that our approach can be used for conditional image generation conditioned on latent features.

arXiv:2505.04146

Unmasking the Canvas: A Dynamic Benchmark for Image Generation Jailbreaking and LLM Content Safety

Existing large language models (LLMs) are advancing rapidly and produce outstanding results in image generation tasks, yet their content safety checks remain vulnerable to prompt-based jailbreaks. Through preliminary testing on platforms such as ChatGPT, MetaAI, and Grok, we observed that even short, natural prompts could lead to the generation of compromising images ranging from realistic depictions of forged documents to manipulated images of public figures. We introduce Unmasking the Canvas (UTC Benchmark; UTCB), a dynamic and scalable benchmark dataset to evaluate LLM vulnerability in image generation. Our methodology combines structured prompt engineering, multilingual obfuscation (e.g., Zulu, Gaelic, Base64), and evaluation using Groq-hosted LLaMA-3. The pipeline supports both zero-shot and fallback prompting strategies, risk scoring, and automated tagging. All generations are stored with rich metadata and curated into Bronze (non-verified), Silver (LLM-aided verification), and Gold (manually verified) tiers. UTCB is designed to evolve over time with new data sources, prompt templates, and model behaviors. Warning: This paper includes visual examples of adversarial inputs designed to test model safety. All outputs have been redacted to ensure responsible disclosure.

arXiv:2505.03844

From Spaceborne to Airborne: SAR Image Synthesis Using Foundation Models for Multi-Scale Adaptation

The availability of Synthetic Aperture Radar (SAR) satellite imagery has increased considerably in recent years, with datasets commercially available. However, the acquisition of high-resolution SAR images in airborne configurations, remains costly and limited. Thus, the lack of open source, well-labeled, or easily exploitable SAR text-image datasets is a barrier to the use of existing foundation models in remote sensing applications. In this context, synthetic image generation is a promising solution to augment this scarce data, enabling a broader range of applications. Leveraging over 15 years of ONERA's extensive archival airborn data from acquisition campaigns, we created a comprehensive training dataset of 110 thousands SAR images to exploit a 3.5 billion parameters pre-trained latent diffusion model \cite{Baqu2019SethiR}. In this work, we present a novel approach utilizing spatial conditioning techniques within a foundation model to transform satellite SAR imagery into airborne SAR representations. Additionally, we demonstrate that our pipeline is effective for bridging the realism of simulated images generated by ONERA's physics-based simulator EMPRISE \cite{empriseem_ai_images}. Our method explores a key application of AI in advancing SAR imaging technology. To the best of our knowledge, we are the first to introduce this approach in the literature.

arXiv:2504.21356공식 코드 · Papers with Code

Nexus-Gen: A Unified Model for Image Understanding, Generation, and Editing

Unified multimodal large language models (MLLMs) aim to integrate multimodal understanding and generation abilities through a single framework. Despite their versatility, existing open-source unified models exhibit performance gaps against domain-specific architectures. To bridge this gap, we present Nexus-Gen, a unified model that synergizes the language reasoning capabilities of LLMs with the image synthesis power of diffusion models. To align the embedding space of the LLM and diffusion model, we conduct a dual-phase alignment training process. (1) The autoregressive LLM learns to predict image embeddings conditioned on multimodal inputs, while (2) the vision decoder is trained to reconstruct high-fidelity images from these embeddings. During training the LLM, we identified a critical discrepancy between the autoregressive paradigm's training and inference phases, where error accumulation in continuous embedding space severely degrades generation quality. To avoid this issue, we introduce a prefilled autoregression strategy that prefills input sequence with position-embedded special tokens instead of continuous embeddings. Through dual-phase training, Nexus-Gen has developed the integrated capability to comprehensively address the image understanding, generation and editing tasks. All models, datasets, and codes are published at https://github.com/modelscope/Nexus-Gen.git to facilitate further advancements across the field.

arXiv:2504.20438

PixelHacker: Image Inpainting with Structural and Semantic Consistency

Image inpainting is a fundamental research area between image editing and image generation. Recent state-of-the-art (SOTA) methods have explored novel attention mechanisms, lightweight architectures, and context-aware modeling, demonstrating impressive performance. However, they often struggle with complex structure (e.g., texture, shape, spatial relations) and semantics (e.g., color consistency, object restoration, and logical correctness), leading to artifacts and inappropriate generation. To address this challenge, we design a simple yet effective inpainting paradigm called latent categories guidance, and further propose a diffusion-based model named PixelHacker. Specifically, we first construct a large dataset containing 14 million image-mask pairs by annotating foreground and background (potential 116 and 21 categories, respectively). Then, we encode potential foreground and background representations separately through two fixed-size embeddings, and intermittently inject these features into the denoising process via linear attention. Finally, by pre-training on our dataset and fine-tuning on open-source benchmarks, we obtain PixelHacker. Extensive experiments show that PixelHacker comprehensively outperforms the SOTA on a wide range of datasets (Places2, CelebA-HQ, and FFHQ) and exhibits remarkable consistency in both structure and semantics. Project page at https://hustvl.github.io/PixelHacker.

arXiv:2504.18391

Fast Autoregressive Models for Continuous Latent Generation

Autoregressive models have demonstrated remarkable success in sequential data generation, particularly in NLP, but their extension to continuous-domain image generation presents significant challenges. Recent work, the masked autoregressive model (MAR), bypasses quantization by modeling per-token distributions in continuous spaces using a diffusion head but suffers from slow inference due to the high computational cost of the iterative denoising process. To address this, we propose the Fast AutoRegressive model (FAR), a novel framework that replaces MAR's diffusion head with a lightweight shortcut head, enabling efficient few-step sampling while preserving autoregressive principles. Additionally, FAR seamlessly integrates with causal Transformers, extending them from discrete to continuous token generation without requiring architectural modifications. Experiments demonstrate that FAR achieves $2.3\times$ faster inference than MAR while maintaining competitive FID and IS scores. This work establishes the first efficient autoregressive paradigm for high-fidelity continuous-space image generation, bridging the critical gap between quality and scalability in visual autoregressive modeling.

arXiv:2504.17789

Token-Shuffle: Towards High-Resolution Image Generation with Autoregressive Models

Autoregressive (AR) models, long dominant in language generation, are increasingly applied to image synthesis but are often considered less competitive than Diffusion-based models. A primary limitation is the substantial number of image tokens required for AR models, which constrains both training and inference efficiency, as well as image resolution. To address this, we present Token-Shuffle, a novel yet simple method that reduces the number of image tokens in Transformer. Our key insight is the dimensional redundancy of visual vocabularies in Multimodal Large Language Models (MLLMs), where low-dimensional visual codes from visual encoder are directly mapped to high-dimensional language vocabularies. Leveraging this, we consider two key operations: token-shuffle, which merges spatially local tokens along channel dimension to decrease the input token number, and token-unshuffle, which untangles the inferred tokens after Transformer blocks to restore the spatial arrangement for output. Jointly training with textual prompts, our strategy requires no additional pretrained text-encoder and enables MLLMs to support extremely high-resolution image synthesis in a unified next-token prediction way while maintaining efficient training and inference. For the first time, we push the boundary of AR text-to-image generation to a resolution of 2048x2048 with gratifying generation performance. In GenAI-benchmark, our 2.7B model achieves 0.77 overall score on hard prompts, outperforming AR models LlamaGen by 0.18 and diffusion models LDM by 0.15. Exhaustive large-scale human evaluations also demonstrate our prominent image generation ability in terms of text-alignment, visual flaw, and visual appearance. We hope that Token-Shuffle can serve as a foundational design for efficient high-resolution image generation within MLLMs.

arXiv:2504.18405

HepatoGEN: Generating Hepatobiliary Phase MRI with Perceptual and Adversarial Models

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays a crucial role in the detection and characterization of focal liver lesions, with the hepatobiliary phase (HBP) providing essential diagnostic information. However, acquiring HBP images requires prolonged scan times, which may compromise patient comfort and scanner throughput. In this study, we propose a deep learning based approach for synthesizing HBP images from earlier contrast phases (precontrast and transitional) and compare three generative models: a perceptual U-Net, a perceptual GAN (pGAN), and a denoising diffusion probabilistic model (DDPM). We curated a multi-site DCE-MRI dataset from diverse clinical settings and introduced a contrast evolution score (CES) to assess training data quality, enhancing model performance. Quantitative evaluation using pixel-wise and perceptual metrics, combined with qualitative assessment through blinded radiologist reviews, showed that pGAN achieved the best quantitative performance but introduced heterogeneous contrast in out-of-distribution cases. In contrast, the U-Net produced consistent liver enhancement with fewer artifacts, while DDPM underperformed due to limited preservation of fine structural details. These findings demonstrate the feasibility of synthetic HBP image generation as a means to reduce scan time without compromising diagnostic utility, highlighting the clinical potential of deep learning for dynamic contrast enhancement in liver MRI. A project demo is available at: https://jhooge.github.io/hepatogen

arXiv:2504.15473

Emergence and Evolution of Interpretable Concepts in Diffusion Models

Diffusion models have become the go-to method for text-to-image generation, producing high-quality images from noise through a process called reverse diffusion. Understanding the dynamics of the reverse diffusion process is crucial in steering the generation and achieving high sample quality. However, the inner workings of diffusion models is still largely a mystery due to their black-box nature and complex, multi-step generation process. Mechanistic Interpretability (MI) techniques, such as Sparse Autoencoders (SAEs), aim at uncovering the operating principles of models through granular analysis of their internal representations. These MI techniques have been successful in understanding and steering the behavior of large language models at scale. However, the great potential of SAEs has not yet been applied toward gaining insight into the intricate generative process of diffusion models. In this work, we leverage the SAE framework to probe the inner workings of a popular text-to-image diffusion model, and uncover a variety of human-interpretable concepts in its activations. Interestingly, we find that even before the first reverse diffusion step is completed, the final composition of the scene can be predicted surprisingly well by looking at the spatial distribution of activated concepts. Moreover, going beyond correlational analysis, we show that the discovered concepts have a causal effect on the model output and can be leveraged to steer the generative process. We design intervention techniques aimed at manipulating image composition and style, and demonstrate that (1) in early stages of diffusion image composition can be effectively controlled, (2) in the middle stages of diffusion image composition is finalized, however stylistic interventions are effective, and (3) in the final stages of diffusion only minor textural details are subject to change.